This study provides two important insights: first, the presence of an alcohol ingestion of > 80 g/d was found to be an independent risk factor for severe CAP and prior ambulatory antimicrobial treatment to be protective; second, the causes of severe pneumonia revealed important differences when compared with our study originating from 1984 to 1987 (3). The most striking changes included a lower frequency of legionellosis and the importance of atypical bacterial as well as viral pathogens that were not tested for in our previous report. On the other hand, S. pneumoniae, and, to a lesser extent, GNEB and P. aeruginosa continued to be leading pathogens.

In this study on risk factors of severe CAP, we decided to rely on a hospital-based design. A methodological shortcoming of this design consists of potential biases originating from referral habits. This must be taken into account when interpreting our findings; for example, the fact that control patients were found to have an increased age and more frequently neurological comorbid illness compared with severe pneumonia patients is most probably due to the fact that younger patients with severe pneumonia were more readily referred to a tertiary care hospital. Moreover, the decision to hospitalize a patient implies at least moderate severity of pneumonia in the majority of cases, precluding the identification of highly prevalent potential risk factors such as COPD, chronic heart failure, or diabetes mellitus. On the other hand, the hospital physician may more confidently rely on risk factors identified in hospital-based designs because the patients studied are more likely to be similar to those he will have to treat.

In studies of severe CAP published so far, COPD, alcoholism, chronic heart disease, and diabetes mellitus were the most common comorbid illnesses (3, 4, 6, 9, 14). In addition, lung disease, bronchial asthma, and heart disease were found to be risk factors of pneumonia in the elderly (15). Therefore, these conditions were expected to be potential factors predisposing to severe CAP. Nevertheless, pulmonary comorbidity, chronic heart disease, liver disease, and diabetes mellitus were not found to be associated with severe CAP, even when specified for COPD, liver cirrhosis, and insulin-dependent diabetes mellitus. This may be a result of the hospital-based approach discussed previously as well as the relatively limited number of patients studied. The presence of renal disease as a risk factor was close to significance. Somewhat surprisingly, only few studies of severe CAP have reported patients with chronic renal failure as a comorbid condition (4, 14).

We found alcoholism (as defined by a daily alcohol intake of ⩾ 80 g/d) to represent an important independent risk factor also for severe CAP. This finding is in line with a previous study from our hospital where high alcohol intake (> 100 g/d in men and > 80 g/d in women) was found to be the only independent risk factor of CAP and also to be significantly associated with death (16). In another study, alcoholism was also described as a risk factor of CAP in the elderly (15). Alcoholism has long been suggested to bear a potential role in the pathogenesis of CAP by reducing alertness and favoring aspiration and, possibly, also by an impairment of several local and systemic host immune mechanisms, especially of neutrophil and lymphocyte functions (17, 18). In the present study, a history of high alcohol intake was not associated with an increased risk of severe CAP. Accordingly, in our previous study, neither the total lifetime dose of alcohol nor a history of alcoholism were identified as risk factors, suggesting that acute effects of alcohol consumption are most important for the development of pneumonia.

There was a trend for treatment with corticosteroids ⩽ 20 mg/d to be associated with severe CAP. This finding suggests that even a low dose of corticosteroids has immunosuppressive effects favoring severe pulmonary infection independent of the underlying disease. By the same token, it at least does not support the hypothesis that preexisting treatment with low doses of corticosteroids may modify favorably the inflammatory response.

S. pneumoniae, GNEB, P. aeruginosa as well as bacteremia failed to be significantly associated with severe CAP, all of which were found to bear adverse prognostic potential in several previous studies (3, 6, 8, 14, 19). This failure may be due to age distribution, the use of hospitalized patients as controls, and the limited number of patients studied. Nevertheless, there was a trend toward a higher incidence of GNEB and P. aeruginosa in severe CAP. Accordingly, typical pathogens (mainly S. pneumoniae, S. aureus, E. coli, and P. aeruginosa) were predominant in nonsurvivors as well as in patients presenting with septic shock.

Prior ambulatory antimicrobial treatment was protective against severe CAP also in multivariate analysis. Because of the limited number of pretreated patients with definite cause of disease, we were not able to establish a significant relation of inappropriate ambulatory treatment and severe CAP. Nevertheless, consistent with our finding, two studies examining prognostic factors of severe CAP have demonstrated the adverse prognostic potential of inappropriate or a delay in appropriate initial antimicrobial treatment (3, 7). Rapid empirical antimicrobial treatment seems to be of paramount importance for the course of CAP. Most probably, the potentially harmful inflammatory response to the underlying pathogens is most effectively limited by an early reduction of the bacterial load, whereas later on the inflammatory cascades may equally run relatively independently of the initial pathogens and, therefore, of antimicrobial treatment (20, 21). Because current antibiotics do effectively reduce the bacterial load within the first 24 h in the presence of a susceptible pathogen, the most important concern is to introduce them as early as possible.

The comparison of current microbial patterns of patients with severe CAP with previous findings from our group (3) must be interpreted with caution because several diagnostic techniques were not applied in the latter series. These include validated sputum cultures, and serology for C. pneumoniae and viruses. Accordingly, the significantly higher incidence of atypical bacterial pathogens is in part explained by the emergence of C. pneumoniae which was not considered during our previous study period. Although this pathogen is usually found in mild to moderate CAP, it has been also found to cause severe CAP (22). The frequency of M. pneumoniae was similar (3% versus 6%). There was, however, also an unexpected high incidence of C. burnetii, a pathogen not detected in our previous study. The reasons for this are unclear. A comparison of the incidence of viral pathogens is also not possible because these pathogens were not investigated in our previous study. Overall, our findings confirm an important role for atypical bacterial and viral pathogens in severe CAP.

Perhaps the most important change is the significantly lower incidence of severe legionellosis. Recent studies originating from Spain have provided some evidence for a generally decreasing incidence of legionellosis (9, 19, 23). In addition, a British follow-up epidemiological study conducted at the same ICU reported only half of cases of severe legionellosis compared with the previous study (16% versus 30%, p = 0.08) (1, 10). A possible explanation for this decreasing incidence of severe legionellosis may be a more widespread early use of macrolides.

As expected, S. pneumoniae continued to be the leading pathogen in severe CAP. However, the emergence of drug- resistant strains has been dramatic, leading to a high incidence in severe CAP as well as in control patients in our present study. The initiation of an appropriate antimicrobial treatment taking into account these changing susceptibility patterns is pivotal, because we recently could show that the outcome is not affected by drug resistance in the presence of an appropriate antimicrobial treatment (24). GNEB and P. aeruginosa continued to be frequent pathogens in the present study (6% and 5%, respectively). This finding is particularly important because there are considerable variations in the reported frequencies of these pathogens in different settings, ranging from 0–34% for GNEB (1, 6-8, 10) and 0–5% for P. aeruginosa (1, 6-9, 14).

An interesting finding was that nearly 10% of patients with severe CAP had mixed etiologies requiring a combination antimicrobial treatment covering additionally atypical bacterial pathogens (including Legionella spp.) or P. aeruginosa, and a further 3% of patients had mixed infections with typical pathogens including drug-resistant strains of S. pneumoniae. Both a broad-spectrum initial antimicrobial treatment and a comprehensive microbial investigation seem mandatory to cope with the problem of mixed infections in severe CAP.

This follow-up epidemiologic study provides evidence that obvious differences in microbial patterns comparing recent studies are only in part caused by differences in local settings. A considerable part of these discrepancies may additionally originate from the period under study, even in the same local setting. Because the recommended empirical approach in the ATS guidelines is based on microbial patterns derived from several epidemiologic surveys (12), it remains particularly important for the clinician to additionally consider current peculiarities of the local setting as well as global and regional trends of changing microbial and susceptibilty patterns.